The Study Of New Noninvasive Predictive Models For Hepatic Portal Hypertension

Background & ObjectivePortal hypertension is an important hallmark in the natural history of liver cirrhosis, and it is the initial consequence of decompensated cirrhosis. The improvement or worsening of portal hypertension parallels that of the disease and it’s closely related with the presence of complications and prognosis. Hepatic venous pressure gradient (HVPG) is the widely acceptable "gold standard" used to quantify the degree of portal hypertension, which helps to assess the risk of decompensation, prognosis and evaluate the pharmacologic treatment. Hepatic vein catheterization is the gold standard technique to measure portal pressure, but in view of its invasiveness and technical requirement, noninvasive predictive models of HVPG have been getting increased attention in recent years.The research aims to investigate the predictive value of basic clinical information for HVPG and establish an effective noninvasive model to predict HVPG score or grade with correlated clinical indexes by statistical methods, which should shed further light on the treatment of hepatic portal hypertension.MethodsClinical data were analyzed in a retrospective manner, including 245 patients with liver cirrhosis who were admitted to our unit from October 2010 to May 2015. According to the inclusion and exclusion criteria,201 patients were included in the study population finally. Routine examinations and HVPG measurement were performed after admission. We collected and sorted out their basic clinical characteristics including age, sex, etiology, past history (hemorrhage, splenectomy or partial splenic embolism), complications (liver cancer, portal vein thrombosis, ascites and hepatic encephalopathy), endoscopic examination (esophageal and gastric varices, EGV), and CT parameters, and other biochemical parameters such as routine blood tests including PLT count, Hb level, ALT level, AST level, CTP score.All the data were analyzed with linear regression analysis, univariate and multivariate logistic regression analysis, Student’s test, Chi-Square test and other regression analysis methods using the statistical software SAS version 9.2 and SPSS 19.0. We applied statistic methods to analyze the correlation between all the above indexes and HVPG, and established new effective noninvasive models to predict HVPG score or grade for closely-correlated clinical indexes, which shows clinical significance (α=0.05,p<0.05).Results1. Multiple liner regression analysis1.1 Single factor analysis:HVPG was significantly correlated with etiology, splenectomy or PSE, portal vein thrombosis, ascites, AST, PLT and CTP score (p=0.0086,0.0403,0.0107, 0.0013, <0.0001,0.0094, <0.0001 respectively),while HVPG showed no significant correlation with age, sex, hemorrhage history, liver cancer, hepatic encephalopathy, ALT, Hb or EGV(p=0.2678,0.2991,0.9854,0.2368,0.4947,0.0625,0.1677,0.774 respectively).1.2 Multi-factor analysis:The stepwise regression method was used to evaluate multiple factors and possible confounding effects. Results indicated that HVPG had remarkable correlation with portal vein thrombosis, AST and CTP score (p= 0.007,0.006, <0.0001 respectively). The predictive model was established as follows.HVPG score=-2.163×(portal vein thrombosis)+0.030×(AST)+0.889×(CTP score)+7.377;[portal vein thrombosis:yes=1, no=0]; R2= 0.2029;The low value of R2 indicated that it was limited to predict HVPG score with this model.2. Establishment of multivariate logistic regression model for HVPG grade:Based on the clinical significance and application, HVPG can be divided to three levels, that is, HVPG level 1(HVPG<12mmHg), HVPG level 2(12mmHg<HVPG <16mmHg) and HVPG level 3(HVPG>16mmHg).The classifications of HVPG and one-way analysis of variance were done by SAS software. HVPG was significantly correlated with AST and CTP score. A new model was established as follows.logitPj=-aj+0.013xAST+0.283x(CTP score) α1=-3.34, a2=-1.7;Correlation analysis of prediction probability and observation response were made to evaluate clinical accuracy of the model (consistency percentage= 70.4, c= 0.706), which showed a good predictive ability. However, the analysis is very complicated showing a limited clinical applicability.3. Models of HVPG grade for some clinical applications:3.1 To assess the risk of variceal bleeding:The patients are divided into two groups, group A (HVPG<12mmHg) and group B(HVPG>12mmHg). The correlation between HVPG and clinical indexes was studied and a predictive model was proposed.(1) Single factor analysisHVPG value was significantly correlated with CTP score, AST, Hb, PLT, etiology, and splenectomy or PSE(p= 0.001,0.002,0.049,0.001,0.002 respectively).(2) Multivariate logistic regression analysis and the predictive model p value=-0.733+0.246×CTP score+0.016xAST-0.001×PLT-0.014×Hb+1.012 × etiology-1.004×splenectomy or PSE (etiology:viral=1, non-viral=0; splenectomy or PSE:yes=1, no=0)(3) ROC curve analysisROC curve was plotted to evaluate the validity of this model, the AUC was 0.773. The cut-off value was 0.6219 with a sensitivity 73.02 and a specificity 71.83, which indicated a high accuracy.3.2 To assess clinical prognosis of liver cirrhosis:HVPG>16mmHg is a good predictor for prognosis of liver cirrhosis, and it is closely correlated with risks of refractory ascites, hepatorenal syndrome, recurrent variceal bleeding and other complications. The patients are divided into two groups, group A with HVPG<16mmHg and group B HVPG>16mmHg. The univariate and multivariate logistic analysis are done.(1) Single factor analysisHVPG value was significantly correlated with CTP score, AST, PUT, and etiology(p<0.001,0.001,0.005,0.043,0.037 respectively).(2) Multivariate logistic regression analysis and the predictive model p value=-2.727+0.231×CTP score+0.014×AST-0.004×PLT+0.416×etiology (etiology:viral=1, non-viral=0)(3) ROC curve analysisThe AUC was 0.744 (p<0.001). The cut-off value was 0.3367 with a sensitivity 80.52 and a specificity 61.67, which indicated a high accuracy.Conclusion1. For patients with hepatic portal hypertension, HVPG value was significantly correlated with etiology, splenectomy or PSE, portal vein thrombosis, ascites, AST, PLT and CTP score. In addition, AST, portal vein thrombosis and CTP score were independent predictors for HVPG value.2. To achieve effective clinical value, new models are developed to predict risks of variceal bleeding and the overall prognosis respectively.(1)p value=-0.733+0.246xCTP score+0.016xAST-0.001×PLT-0.014×Hb+1.012 ×etiology-1.004xsplenectomy or PSE(2) p value=-2.727+0.231 xCTP score+0.014xAST-0.004xPLT+0.416xetiology (etiology:viral=1, non-viral=0; splenectomy or PSE:yes=1, no=0)Both AUC of these two predictive models were among 0.7-0.9, which indicated a good accuracy.